Wavelet Mean Reversion Server

Overview

This server decomposes a numeric series into wavelet scales, analyzes mean-reversion properties, generates signals, and provides simple backtests and visualizations. Supplied and synthetic data work without an external provider; Yahoo retrieval requires network access.

Connection

MCP URL: https://quantx-api.limex.pro/mcps/wavelet-mean-reversion Select Wavelet Mean Reversion in the QuantX marketplace. Use its gateway connection and QuantX authorization. See Client Setup. Cursor configuration (.cursor/mcp.json):

{
  "mcpServers": {
    "wavelet-mean-reversion": {
      "url": "https://quantx-api.limex.pro/mcps/wavelet-mean-reversion",
      "headers": {
        "Authorization": "Bearer ${env:QUANTX_MCP_TOKEN}"
      }
    }
  }
}

Set the referenced environment variables before starting Cursor. For Claude Code or Python, use the client-specific format in Client Setup.

Tools

Data and Analysis

Tools	Purpose
`yahoo-data-fetcher-fetch`	Retrieve one ticker’s prices; `ticker` must be a string, such as `"SPY"`
`yahoo-data-fetcher-fetch-returns`	Retrieve returns for one string `ticker`
`yahoo-data-fetcher-fetch-multiple-series`	Retrieve several series using `tickers`, such as `["SPY", "QQQ"]`
`synthetic-data-generator-generate-mean-reverting`, `synthetic-data-generator-generate-trending`, `synthetic-data-generator-generate-complex-signal`, `synthetic-data-generator-add-noise`	Generate controlled synthetic input
`validate-time-series`	Validate a one-dimensional Series object with at least four finite values; see the constructor example below
`wavelet-analyzer-decompose`, `wavelet-analyzer-analyze`	Decompose scales and produce analysis
`wavelet-decomposition-result-get-scale-signal`	Extract a scale from a decomposition object
`wavelet-analysis-result-summary`, `wavelet-analysis-result-get-statistics-dataframe`	Summarize an analysis object or extract its statistics

Signals, Backtests, and Charts

Tools	Purpose
`wavelet-mean-reversion-strategy-fit`, `wavelet-mean-reversion-strategy-generate-signals`	Fit/analyze input and generate signals
`trading-signal-to-dataframe`	Convert a signal object into a table
`wavelet-mean-reversion-strategy-backtest-simple`	Run a simple backtest using the required strategy/signal inputs
`compare-with-bollinger-bands`	Compare the strategy with a Bollinger-band baseline
`wavelet-visualizer-plot-trading-signals`, `wavelet-visualizer-plot-backtest-results`	Plot signals or a backtest
`wavelet-visualizer-plot-complete-analysis`, `wavelet-visualizer-plot-acf-analysis`, `wavelet-visualizer-plot-monte-carlo-results`	Plot analysis, ACF, or simulation results

Configuration and Object Lifecycle

config-from-dict and config-to-dict convert configuration. The presets are default-config, monte-carlo-default, statistical-analyzer-default, wavelet-analyzer-default, wavelet-strategy-default, wavelet-visualizer-default, and yahoo-fetcher-default. Each of these object prefixes has -create, -info, -list, and -delete tools:

config
yahoo-data-fetcher
trading-signal
wavelet-mean-reversion-strategy
wavelet-visualizer
wavelet-analysis-result
wavelet-analyzer
wavelet-decomposition-result

wavelet-mean-reversion-strategy-create requires an analyzer. The shortest path is wavelet-strategy-default with {"request": {}}, which returns the strategy handle and deps.analyzer. The demonstration entry points are main-main, example-synthetic-data, example-real-market-data, example-monte-carlo-backtest, example-multiple-tickers, and example-custom-configuration. Inspect their schemas and expected cost before using them in automation. Do not infer additional callable methods from a class name.

Examples

Retrieve One or Several Tickers

Call yahoo-data-fetcher-fetch with a single string ticker:

{
  "request": {
    "ticker": "SPY",
    "start_date": "2026-01-01",
    "end_date": "2026-02-01",
    "interval": "1d"
  }
}

For more than one ticker, call yahoo-data-fetcher-fetch-multiple-series. The field is tickers, not ticker; select the price field with column. This tool has no interval parameter:

{
  "request": {
    "tickers": ["SPY", "QQQ"],
    "start_date": "2026-01-01",
    "end_date": "2026-02-01",
    "column": "Close"
  }
}

Both calls depend on Yahoo availability. Passing an array to the single-ticker tool fails schema validation even if older descriptions suggest otherwise.

Generate a Synthetic Series

Call synthetic-data-generator-generate-mean-reverting:

{
  "request": {
    "N": 512,
    "theta": 0.1,
    "mu": 0.0,
    "sigma": 1.0,
    "seed": 42
  }
}

Resolve a returned result_id and extract the numeric series. Analysis tools take those numbers as signal. validate-time-series is the exception: it rejects a plain JSON array and needs the values wrapped in the Series object shown below.

Validate a Small Input

Call validate-time-series with a nested pandas.Series constructor. data here is an object specification, not a raw list:

{
  "request": {
    "data": {
      "__class__": "pandas:Series",
      "init_kwargs": {
        "data": [0.2, -0.1, 0.15, -0.05, 0.1, -0.08, 0.04, -0.02],
        "dtype": "float64"
      }
    }
  }
}

Four points is only the validator’s minimum. A useful wavelet analysis requires enough history for the chosen wavelet and decomposition level; do not interpret this short example as research evidence.

Pass Object Handles

Create the required objects with -create or their preset tools and inspect the returned handle. A receiver uses handle, while a nested object parameter uses {"__handle__": "<handle>"}. wavelet-analyzer-analyze and wavelet-mean-reversion-strategy-generate-signals return JSON payloads rather than object handles. That limits chaining: wavelet-mean-reversion-strategy-backtest-simple and the visualizers expect a TradingSignal or WaveletAnalysisResult object, so read the returned signal JSON yourself instead of feeding it to them. For wavelet-mean-reversion-strategy-generate-signals, supply the strategy handle and the input signal. Set refit when the strategy should fit the new series instead of reusing its previous fit; check the current schema for the field name.

Results and Limitations

Analysis comes back either as structured JSON or as an object handle. For a handle, read it with the matching tool — wavelet-analysis-result-summary or wavelet-analysis-result-get-statistics-dataframe. Large results are available through result_get. Charts return image data with artifact_type: "image" and mime: "image/png", often inside a stored result. NaN/Inf values are rejected. Save returned analysis separately. A full-series decomposition or in-sample backtest is not automatically causal or out of sample; use time-separated evaluation for research claims. See Shared MCP Tools for artifacts and object handling.

Overview

MCP Setup

Dataloader Server

Fama-French Replicate Server

Statistical Factor Models Server

Jump Models Server